KazNERD2

Synthetic Kazakh speech dataset from the paper Improved Kazakh Named Entity Transcription Using Synthetic Speech (UBMK 2025). It contains TTS-synthesized speech of modified KazNERD training sentences (the transcripts differ from the original KazNERD text) and is used as additional training data.

Code and evaluation scripts: github.com/IS2AI/KazNERD_ASR

Related datasets: issai/kaznerd1 (synthetic speech of the original sentences), issai/KazNERD_H (human-recorded test set)

Dataset Structure

Split	Utterances
train	35,609

Data Fields

• audio: 16 kHz mono WAV audio
• text: Kazakh transcript
• sentence_id: ID of the KazNERD sentence it was derived from, without its trailing AID (e.g. AID343740 ↔ AID343740AID)
• speaker_id: synthetic speaker ID
• gender: speaker gender (male / female)
• tts_model: TTS model used to generate the audio

Usage

from datasets import load_dataset

dataset = load_dataset("issai/kaznerd2", split="train")
sample = dataset[0]
print(sample["text"], sample["tts_model"], sample["speaker_id"])

Citation

@inproceedings{meiramov2025kaznerdasr,
  author    = {Meiramov, Rakhat and Varol, Huseyin Atakan},
  title     = {Improved Kazakh Named Entity Transcription Using Synthetic Speech},
  booktitle = {2025 10th International Conference on Computer Science and Engineering (UBMK)},
  pages     = {1629--1633},
  year      = {2025},
  doi       = {10.1109/UBMK67458.2025.11242232}
}

Transcripts are based on KazNERD; please also cite:

@inproceedings{yeshpanov-etal-2022-kaznerd,
  author    = {Yeshpanov, Rustem and Khassanov, Yerbolat and Varol, Huseyin Atakan},
  title     = {{K}az{NERD}: {K}azakh Named Entity Recognition Dataset},
  booktitle = {Proceedings of the Thirteenth Language Resources and Evaluation Conference},
  pages     = {417--426},
  year      = {2022},
  url       = {https://aclanthology.org/2022.lrec-1.44}
}